Recently, the amount of data obtained from astronomical instruments has been increasing explosively, and data science methods such as Machine Learning/Deep Learning gain attention on the back of the growth in demand for automatic analysis. Using these methods, the number of applications to the target sources that have clear boundaries with the background i.e., stars, planets, and galaxies is increasing year by year. However, there are a few studies which applied the data science methods to the interstellar medium (ISM) distributed in the Galactic plane, which have complicated and ambiguous silhouettes. We aim to develop classifiers to automatically extract various structures of the ISM by Convolutional Neural Network (CNN) that is strong in image recognition even in deep learning. In this study, we focus on the infra-red (IR) ring structures distributed in the Galactic plane. Based on the catalog of Churchwell et al. (2006, 2007), we created a "Ring"dataset from the Spitzer/GLIMPSE 8 μm and Spitzer/MIPSGAL 24 μm data and optimized the parameters of the CNN model. We applied the developed model to a range of 16.5° ≤ l ≤ 19.5°, |b| ≤ 1°. As a result, 234 "Ring"candidates are detected. The "Ring"candidates were matched with 75%Milky Way Project (MWP, Simpson et al. 2012) "Ring"and 65%WISE Hii region catalog (Anderson et al. 2014). In addition, new"Ring"and Hii region candidate objects were also found. For these results, we conclude that the CNN model may have a recognition accuracy equal to or better than that of human eyes.